Analytics in Mineral Processing

Introduction

Mineral processing plants are one of the most data rich parts of any mining operation.  Data is generated and stored for practically every step of the process but do we get the most value from this data?

Historically, process analysis and optimisation has been undertaken by skilled operators and metallurgists, experienced in monitoring fluctuations in behaviour.  However, this relies heavily on the presence of highly skilled people with sufficient experience in the operation to interpret results. Data from individual unit operations is often used in this process to support hypotheses but the capability to use integrated data sets from across the operations has not been available.

 

The development of advanced data mining techniques, deep learning, machine learning and artificial intelligence now means that the tools are available in mineral processing to better use our data.  However, while uptake of advanced analytics has accelerated over the past few years in exploration, mining and geometallurgy there has yet to be a revolutionary advance in how we use data in the mineral processing plant.

 

In recent years advances have been made in development and implementation of tools such as Digital Twins and today we will explore how those can be used to drive genuine improvements.

 

The problem

The major hurdle for advanced analytics in mineral processing is the complexity and variability of data collected.  Data is often available on different time series, ranging from milliseconds for sensors to monthly for composite samples.  This makes correlations between datasets complex, requiring additional levels of computation. Add to this the requirement to track the material through the process so that relationships between unit operations can be established.

 

Combined these issues are not dissimilar to what is experienced in the manufacturing industry, where they have been broadly solved.  Where mineral processing is unique from manufacturing is that the feedstock is generally highly variable and often unknown.

 

The solution

Great advances have been made in the development of Digital Twins and ‘soft sensors’.  A Digital Twin provides a dynamic simulation of the process, allowing prediction of process behavior based on historical or current data.  This can be used to test the impact of process changes or fill in sparse data, such as a tailings grade. The latter often referred to as a ‘soft sensor’  A great summary of Digital Twins in mining can be found at the Intelligent Miner.

 

Several examples of Digital Twins are in use today, including;

 

• Andritz IDEAS Digital Twin – The IDEAS simulation software is well established in steady state simulation of minerals processing operations.  It is a diverse platform with libraries covering many industries and Andritz has been a leader in development of Digital Twins for mineral processing plants.  The application has successfully targeted building soft sensors for difficult to measure parameters.
• Petra Datascience  MAXTA – PETRA Datascience is a great example of the new wave of analytics focused companies helping operations get more from their data.  The MAXTA product was developed for geometallurgical and mine-to-mill operations but they have demonstrated how this can be extended into the minerals processing plant.  Using the concept that the mill is the best laboratory they can track material from the mine plan to the processing plant, assigning behaviour characteristics by block. This can then be fed back to the mine plan, allowing prediction of key parameters like recovery much earlier in the process.
• GE Digital Mine –  The GE Digital Mine framework is based on their Predix platform for the industrial internet.  It is an all encompassing solution to make data more accessible to operators and drive data driven decision making. 
• Metallurgical Systems – Met Systems have developed a solution for real time monitoring of processing by reading tags within the local historian and modelling the process response.  This allows soft sensors to be developed and gives operators access to data they wouldn’t normally be able to see.

 

Case studies are beginning to emerge for operations that have implemented these technologies and seen great improvements.  Petra Datascience published a case study for the integration of mine planning and mineral processing at the Ban Houayxai gold mine.  This is one of the best examples of linking process behavior to the resource block model or mine plan to define how ore types should behave.  

 

Andritz have published a case study showing how virtual instrumentation can be used to improve efficiency at the OceanaGold Haile operation.  This implementation of a digital twin began with the hydrocyclone circuit and has been expanded across the plant.

 

Met Systems have published a number of case studies on implementation of their solutions.

 

The common goal of digital twin implementations is to reduce variability and achieve improvements that way.  This has been demonstrated many times to result in immediate improvements and cost savings for most operations.

 

The McKinsey MineLens solution looks to add another dimension to this by providing a benchmark between operations.  Allowing management to understand where the operations sits in relation to its peers. The case study at Bagdad Copper Mine is a great example of how this implementation has occurred. 

 

Bringing this one step forward, integrating a Digital Twin with unit operation models can allow the behavior to be predicted for unknown ore types.  This can be used in process development to improve the efficiency and effectiveness at feasibility level. We recently used this approach in the Definitive Feasibility Study for Aura Energy’s Tiris Uranium Project, greatly reducing the engineering and variability test work requirements.

 

The real power of advanced analytics for mineral processing is still to be achieved.  In our opinion the next step should be to combine the Digital Twin with an Integrated Modelling Framework and cost driven decision model.  This will allow operations to assess the impact of ore variability or process changes on a cost basis, incorporating all parts of the process and not just single unit operations or circuits in isolation.  This will lead to often unanticipated cost and productivity benefits that could not be achieved by looking deeper into the data.

 

If you want more information on analytics in mineral processing feel free to contact us at any time